Apple’s Mac Studio with the M5 Ultra chip has landed in front of one of the toughest graders in tech journalism, and it came out on top. In a review published September 24, 2026, Ars Technica’s Andrew Cunningham put the new machine through weeks of benchmarks and daily use, and the conclusion was blunt: this is the quickest chip Apple has ever shipped. The review lands almost exactly a month after Apple’s August 25 announcement of the machine, giving the market its first independent read on whether the price jump to $12,299 for a loaded configuration is actually justified.
Ars Technica Calls the M5 Ultra “The Fastest Apple Silicon Processor Built So Far”
Cunningham’s headline finding sets the tone for the entire piece, as reported by MacDailyNews‘ coverage of the review.
“The M5 Ultra is, by almost every measure, the fastest Apple Silicon processor built so far.”
Andrew Cunningham, Senior Technology Writer, Ars Technica
That statement carries weight because Ars Technica has reviewed every Apple Silicon chip since the transition away from Intel began in 2020. Cunningham’s testing covered both raw CPU throughput and GPU compute, plus real-world workloads like video encoding and local AI inference. The review lands the M5 Ultra squarely at the top of Apple’s internal chip hierarchy, ahead of the M5 Max that ships in the MacBook Pro and Mac Studio’s standard tier, and ahead of every previous Ultra-class chip Apple has built.
Buyers searching for an M5 Ultra Mac Studio review now have a credible third-party benchmark to weigh against Apple’s own marketing claims, which is exactly the kind of validation that shapes purchasing decisions for studios, developers, and AI researchers deciding whether to spend five figures on a desktop.
What Apple Announced on August 25
Apple introduced the M5 Ultra Mac Studio on August 25, 2026, positioning it as the machine for creators, engineers, and AI developers who need desktop-class compute without renting cloud GPUs. The base configuration ships with a 30-core CPU, a 64-core GPU, 96GB of unified memory, and 1TB of storage, priced at $5,499 in the United States (or $5,099 for education buyers). At the top end, Apple lists configurations up to a 36-core CPU, an 80-core GPU, and 512GB of unified memory, a jump that lets a single desktop hold models that would otherwise require a multi-GPU server rack.
Apple’s own newsroom announcement leaned hard into the AI angle rather than traditional creative workloads like video editing or 3D rendering, which had been the Mac Studio’s original pitch back in 2022.
“With the powerful M5 Ultra, Mac Studio scales up to a 36-core CPU, up to an 80-core GPU, and a staggering 512GB of unified memory, enabling users to run enormous LLMs entirely on device.”
Apple, official company statement, Apple Newsroom
That framing matters. Apple is no longer selling the Mac Studio purely on editing timelines or export speeds. It is selling unified memory capacity as the feature that lets a $12,299 desktop do something a $30,000 GPU server cluster does, at least for inference on large models that fit inside 512GB of RAM.
Mac Studio M5 Ultra Pricing: $5,499 to $12,299
Pricing is where most buyers will actually make their decision, and the spread is wide. The entry configuration starts at $5,499 through Apple’s online store. Cunningham’s review unit, by contrast, was configured with a 36-core CPU, an 80-core GPU, 256GB of unified memory, and 4TB of storage, which brought the as-tested price to $12,299. That’s more than double the entry price for roughly the core-count ceiling minus half the maximum memory.
| Configuration | CPU Cores | GPU Cores | Unified Memory | Storage | Price (USD) |
|---|---|---|---|---|---|
| Entry (Education) | 30-core | 64-core | 96GB | 1TB | $5,099 |
| Entry (Standard) | 30-core | 64-core | 96GB | 1TB | $5,499 |
| As-tested (Ars Technica review unit) | 36-core | 80-core | 256GB | 4TB | $12,299 |
| Maximum listed configuration | 36-core | 80-core | 512GB | Not disclosed | Not disclosed |
Apple has not published a public price for the full 512GB configuration paired with maximum storage, so treat that figure as directionally higher than $12,299 rather than a confirmed number. What is confirmed is that jumping from the entry tier to Cunningham’s review configuration costs an extra $6,800, mostly for the memory and storage upgrades rather than the CPU and GPU core counts, which only moved from 30 to 36 cores and 64 to 80 cores respectively.
Under the Hood: M5 Ultra Specs Breakdown
The M5 Ultra’s headline specs read like a spreadsheet built for AI workloads first and everything else second. Beyond the CPU and GPU core counts, Apple confirmed a 32-core Neural Engine and 1.2TB/s of memory bandwidth, both aimed squarely at machine learning throughput rather than gaming or general desktop responsiveness.
The Ultra tier has always come from fusing two Max-class dies into a single package, a design Apple has used since the original M1 Ultra debuted in 2022. That approach is what lets the chip scale to 36 CPU cores and 80 GPU cores without a second, physically separate processor. It also explains why the M5 Ultra inherits most of its per-core architecture from the M5 Max rather than introducing a fundamentally new design, which is part of why the comparisons against M5 Max are just as revealing as the comparisons against the previous-generation M3 Ultra.
For readers tracking Apple’s broader silicon roadmap, the M5 Ultra’s Neural Engine and memory bandwidth numbers sit alongside recent moves on the mobile side, including the A20 Pro’s on-device AI performance gains, suggesting Apple is pushing local inference capability across its entire silicon lineup rather than just the desktop flagship.
Benchmark Results: M5 Ultra vs. M3 Ultra
The most direct generational comparison in Cunningham’s testing pits the M5 Ultra against its immediate predecessor for this tier, the M3 Ultra. Across Ars Technica’s benchmark suite, the M5 Ultra posted roughly 30% faster single-core CPU performance and about 50% faster multi-core CPU performance than the M3 Ultra. GPU gains varied by test, landing between 33% and 66% faster depending on the workload.
| Metric | M5 Ultra vs. M3 Ultra | M5 Ultra vs. M5 Max |
|---|---|---|
| Single-core CPU | ~30% faster | Included in general CPU gain below |
| Multi-core CPU | ~50% faster | 80%-90% faster |
| GPU (general tests) | 33%-66% faster | 50%-80% faster |
| Peak AI compute (Apple figures) | 4.3x faster | Not disclosed by Apple |
That 50% multi-core gain is a meaningful jump for a chip that, on paper, only added six CPU cores over its predecessor. It points to per-core architectural improvements carried over from the M5 generation’s mobile and laptop chips, rather than the gain coming purely from added core count. It’s a pattern readers have already seen play out on the GPU side too, echoing the kind of generational leap covered in our look at how the Apple M6 GPU closed the gap with the RTX 4080.
The Bigger Gap: M5 Ultra vs. M5 Max
The more striking comparison in the review isn’t against last generation’s Ultra chip. It’s against this generation’s Max chip, the one that powers the standard Mac Studio tier and the top MacBook Pro. Cunningham found the gap between M5 Ultra and M5 Max to be far wider than the gap between generations.
“In many of our general-purpose CPU and GPU tests, the Ultra’s CPU outruns the M5 Max by 80 or 90 percent, and the GPU is between 50 and 80 percent faster.”
Andrew Cunningham, Senior Technology Writer, Ars Technica
An 80 to 90 percent CPU lead over the Max chip, nearly double in some tests, is a bigger jump than the entire generational leap from M3 Ultra to M5 Ultra. That math makes the Ultra tier look far more compelling for CPU-bound work than the price gap alone would suggest, since buyers aren’t just paying for a modest bump, they’re paying for what is functionally two Max chips working as one.
The One Chip That Still Beats It: Mac Mini’s M6
Not every benchmark went the Ultra’s way, and Cunningham was specific about the exception. The Mac mini’s M6 chip, a far cheaper and smaller machine, narrowly edged out the M5 Ultra in single-core CPU performance. That’s not a contradiction of the “fastest ever” headline claim so much as a reminder that single-core performance and total system throughput are different measurements, and Apple’s newest small-form chip happens to have a slight edge in raw per-core clock efficiency.
It’s a nuance worth sitting with. A machine costing a fraction of $12,299 can still out-sprint the flagship on single-threaded tasks, even while losing badly on every multi-core and GPU-heavy test. Readers who followed the earlier Geekbench leaks comparing M5 Ultra and M6 scores before this review published will recognize the pattern: raw core count and unified memory capacity win the workstation argument, but they don’t automatically win every single benchmark line item.
Cunningham’s Verdict: Who Should Actually Buy This
Speed numbers aside, the review’s most useful section for buyers is Cunningham’s read on who the machine is actually for. His answer is narrower than Apple’s marketing might suggest.
“The Ultra version of the Mac Studio remains a niche product for a small audience: People with thousands of dollars to spend and who want to work with AI models, who are OK with good-but-not-cutting-edge speeds, and who don’t want to pay OpenAI or Anthropic or whoever for all the tokens they might burn with playful experiments or by absorbing and editing a large codebase…”
Andrew Cunningham, Senior Technology Writer, Ars Technica
That’s a pointed framing. Cunningham isn’t describing a machine for video editors chasing export speed or 3D artists rendering scenes overnight, the audience Apple originally built the Mac Studio for back in 2022. He’s describing developers and AI tinkerers weighing the Mac Studio’s up-front cost against a recurring API bill from a cloud AI provider. It’s a cost-of-ownership argument as much as a performance one, and it reframes the M5 Ultra less as a workstation chip and more as a hedge against token pricing.
Local AI Workloads: Why Unified Memory Matters Now
The unified memory ceiling is the actual story here, more than raw clock speed. A 512GB pool shared between CPU and GPU means large language models that would otherwise need to be split across multiple discrete GPUs can instead run entirely inside one machine’s memory. Apple’s own AI compute figures back up why that matters for the company’s pitch to developers.
“With M5 Ultra, Mac Studio achieves up to 4.3x the peak AI compute performance of M3 Ultra and a staggering 9.8x more than M1 Ultra.”
Apple, official company statement, Apple Newsroom
Cunningham’s own daily-use impressions echo why that ceiling matters beyond the benchmark charts.
“It’s genuinely fun and freeing to be able to write hobby-project code at usable speeds without my data ever leaving my control.”
Andrew Cunningham, Senior Technology Writer, Ars Technica
That privacy and control angle is going to matter more to some buyers than the raw speed numbers. Running a coding assistant or an experimental model locally means nothing about that session leaves the machine, a different value proposition than subscribing to a hosted API where every prompt and every line of code passes through a third party’s servers.
Historical Context: Apple’s Ultra Chip Lineage
Apple’s Ultra-tier chips trace back to the M1 Ultra, the design that introduced fusing two Max dies into a single package for the original Mac Studio launch in 2022. The M2 Ultra and M3 Ultra followed in successive Mac Studio refreshes, each roughly doubling core counts and memory bandwidth over its predecessor while keeping the same fundamental dual-die approach. The M5 Ultra continues that lineage, and Apple’s own comparison figures, a 4.3x peak AI compute gain over M3 Ultra and 9.8x over the original M1 Ultra, give a sense of how far the architecture has scaled across four hardware generations.
For more on how Apple Silicon’s architecture evolved from the original M1 through to today’s chips, Wikipedia’s Apple silicon overview tracks the broader chip family history. What’s changed most between generations isn’t the fusion concept itself, it’s the memory ceiling and the AI-specific compute Apple keeps layering on top of the same basic idea.
Competitive Landscape: Mac Studio vs. AI Workstation PCs
The M5 Ultra doesn’t compete in a vacuum. Windows and Linux workstation builders have spent the last two years chasing the same local-AI buyer with discrete GPU setups, and the pitch on that side of the market usually comes down to raw VRAM stacked across multiple cards versus Apple’s single unified pool. Machines like System76’s Thelio Mira AI, which undercuts Nvidia’s own DGX Spark on price, represent the PC side of that argument: cheaper entry points, but memory that’s split across separate GPU dies rather than shared as one contiguous pool.
That distinction is exactly why Cunningham’s framing about avoiding per-token cloud AI costs resonates. A 512GB unified memory pool sidesteps the complexity of sharding a large model across multiple GPUs, at the cost of losing the raw parallel throughput a stacked-GPU rig can offer for training rather than inference. For inference-heavy, single-user workloads, the Mac Studio’s simplicity is arguably the bigger selling point than any individual benchmark score.
Market Impact: Apple’s Pro Strategy Bet
Apple’s decision to lean on AI compute messaging for the Mac Studio launch, rather than the video and audio production framing it used in 2022, signals a shift in who the company thinks its highest-margin desktop buyer actually is. The Mac Pro tower has struggled to find a clear identity since the Apple Silicon transition, and the Mac Studio has effectively absorbed the role of Apple’s top-end workstation. Pricing the entry M5 Ultra configuration at $5,499, a jump from where the M3 Ultra tier started, suggests Apple believes buyers willing to spend that much are already thinking in terms of AI infrastructure cost, not just creative software licenses.
It also puts pressure on the rest of Apple’s desktop lineup to justify its own AI credentials. The macOS 27 Golden Gate release, which drops all Intel Mac support, and the broader software push around on-device AI features both point toward Apple treating Apple Silicon’s memory architecture as a first-class differentiator against Windows PCs, not just a battery-life advantage the way Apple pitched it during the original M1 launch.
Which Configuration Should You Actually Buy
Given the confirmed pricing tiers listed on Apple’s official Mac Studio specs page, the decision mostly comes down to memory needs rather than core counts. The jump from 30 to 36 CPU cores and 64 to 80 GPU cores between the entry and top configurations is real but modest next to the generational gains Cunningham measured against M3 Ultra and M5 Max. Memory capacity is what actually changes what kind of model you can run locally.
- Buyers doing general creative work or standard development: the $5,499 entry configuration with 96GB of memory covers most non-AI workloads without paying for capacity that will sit unused.
- Developers running mid-size local models or coding assistants: Cunningham’s $12,299 as-tested configuration, with 256GB of memory, represents a reasonable middle ground.
- Researchers and studios running the largest open-weight models entirely on device: the 512GB maximum configuration is the only option, though Apple hasn’t published its exact price.
Anyone who doesn’t have a specific local-AI use case in mind is likely better served by the entry tier or by the standard M5 Max Mac Studio, given how narrow Cunningham described the ideal buyer profile for the Ultra chip specifically.
What Comes Next: Predictions for Apple Silicon
A handful of trends look likely to follow from this review cycle:
- Expect Apple’s marketing for future Mac Studio refreshes to keep leading with AI compute and unified memory capacity rather than video export benchmarks, following the pattern set by this launch.
- The gap Cunningham found between M5 Ultra and M5 Max, wider than the gap between Ultra generations, suggests Apple will keep widening Max-to-Ultra differentiation to justify the Ultra tier’s price premium.
- Competing PC workstation builders will likely respond with even larger unified or pooled-memory configurations, since the “run it all on one machine without cloud tokens” argument is resonating with developers.
- Software makers building local AI tools and coding assistants have new incentive to optimize specifically for Apple Silicon’s memory bandwidth, given how directly Cunningham tied his own workflow to the chip’s local inference speed.
- Pricing for the top 512GB configuration, still undisclosed, will likely land well north of $15,000 once Apple publishes full build-to-order pricing, based on how the jump from 96GB to 256GB already added $6,800 to the base price.
Frequently Asked Questions
What is the starting price of the Apple M5 Ultra Mac Studio?
The entry configuration starts at $5,499 in the United States, or $5,099 for education buyers, and includes a 30-core CPU, a 64-core GPU, 96GB of unified memory, and 1TB of storage.
How much faster is the M5 Ultra than the M3 Ultra?
According to Ars Technica’s Andrew Cunningham, the M5 Ultra delivers roughly 30% faster single-core CPU performance, about 50% faster multi-core CPU performance, and 33% to 66% faster GPU performance compared with the M3 Ultra, depending on the specific test.
Is the M5 Ultra actually faster than the M5 Max?
Yes, and by a wider margin than the generational comparison. Cunningham found the M5 Ultra’s CPU outperformed the M5 Max by 80% to 90% in general-purpose tests, while its GPU ran 50% to 80% faster.
Did any chip beat the M5 Ultra in testing?
Yes. The Mac mini’s M6 chip narrowly beat the M5 Ultra in single-core CPU performance, though the M5 Ultra remained well ahead in multi-core and GPU-heavy tests.
How much unified memory can the M5 Ultra Mac Studio support?
Apple lists a maximum configuration of 512GB of unified memory, alongside a 36-core CPU and an 80-core GPU. Apple has not published a public price for that exact top-tier build.
Who reviewed the Apple M5 Ultra Mac Studio?
Andrew Cunningham, a senior technology writer at Ars Technica, published the review on September 24, 2026, testing a configuration priced at $12,299.
Is the Mac Studio M5 Ultra good for running AI models locally?
Apple says the chip delivers up to 4.3 times the peak AI compute performance of the M3 Ultra and 9.8 times that of the original M1 Ultra, and Cunningham specifically praised running local coding tools and experimental models on the machine without sending data to a third-party cloud service.
When did Apple announce the M5 Ultra Mac Studio?
Apple announced the Mac Studio with M5 Ultra on August 25, 2026. The independent Ars Technica review followed exactly one month later, on September 24, 2026.
